RC22837 (W0307-028) July 2, 2003 Mathematics IBM Research Report Stable Ergodicity Charles Pugh University of California at Berkeley Berkeley, CA 94720 Michael Shub IBM Research Division Thomas J. Watson Research Center P.O. Box 218 Yorktown Heights, NY 10598 Research Division Almaden - Austin - Beijing - Delhi - Haifa - India - T. J. Watson - Tokyo - Zurich LIMITED DISTRIBUTION NOTICE: This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g. , payment of royalties). Copies may be requested from IBM T. J. Watson Research Center , P. O. Box 218, Yorktown Heights, NY 10598 USA (email: [email protected]). Some reports are available on the internet at http://domino.watson.ibm.com/library/CyberDig.nsf/home . STABLE ERGODICITY CHARLES PUGH AND MICHAEL SHUB June 30, 2003 1. Introduction A dynamical system is ergodic if it preserves a measure and each measurable invariant set is a zero set or the complement of a zero set. No measurable invariant set has intermediate measure. See also Section 6. The classic real world example of ergodicity is how gas particles mix. At time zero, chambers of oxygen and nitrogen are separated by a wall. When the wall is removed the gasses mix thoroughly as time tends to innity. In contrast think of the rotation of a sphere. All points move along latitudes, and ergodicity fails due to existence of invariant equatorial bands. Ergodicity is stable if it persists under perturbation of the dynamical system. In this paper we ask: “how common are ergodicity and stable ergodicity?” and we propose an answer along the lines of the Boltzmann hypothesis – “very.” There are two competing forces that govern ergodicity – hyperbolicity and the KAM phenomenon. The former promotes ergodicity and the latter impedes it. One of the striking applications of KAM theory and its more recent variants is the existence of open sets of volume preserving dynamical systems each of which possesses a positive measure set of invariant tori and hence fails to be ergodic. Stable ergodicity fails dramatically for these systems. But does the lack of ergodicity persist if the system is weakly coupled to another? That is, what happens if you have a KAM system or one of its perturbations that refuses to be ergodic, due to these positive measure sets of invariant tori, but somewhere in the universe there is a hyperbolic or partially hyperbolic system weakly coupled to it? – does the lack of egrodicity persist? The answer is “no,” at least under reasonable conditions on the hyperbolic factor. See Section 13 for more details and the proof of Theorem A . If a volume preserving hyperbolic system with suciently strong hyperbolicity is weakly coupled to a KAM system then often the resulting dynamical system is not only ergodic, it is stably ergodic. In short, Hyperbolicity trumps KAM, and ergodicity often reignsa la Boltzmann. Our theme has long been that “a little hyperbolicity goes a long way toward ergodicity,” and we continue to hold hope for the Main Conjecture. Among the volume preserving partially hyperbolic dynamical systems, the stably ergodic ones form an open and dense set. The rst author was supported in part by IBM. The second author was supported in part by NSF Grant #DMS-9988809. 1 2 C. PUGH AND MICHAEL SHUB Theorem A is a step in that direction. See also Section 18. In Section 7 we formulate Theorem E in which we give sucient conditions for ergodicity and stable ergodicity of a partially hyperbolic system. We also discuss at length some examples of stable ergodicity that arise in Lie group dynamics. We are interested in the following question: does stability of ergodicity in which the perturbations occur only in the Lie group context imply stable ergodicity? That is, is ane stable ergodicity decisive for general stable ergodicity? In this direction, we have Theorem B. In a large class of ane dieomorphisms of homogeneous spaces, stable ergodicity within the class is decisive for general stable ergodicity. See Section 16 for details and the proof. Notes. We assume for the most part that the phase space of our dynamical system is a smooth compact manifold M and that m is a smooth invariant volume form 2 1+ on M. Smooth means C∞, although C , or in some cases C with small >0, is usually good enough. Questions about ergodicity in the C1 world seem to be fundamentally dierent. In the case of a ow, for all t R, ϕt : M M is a dieomorphism and ϕt+s = ∈ → ϕt ϕs. In the case of discrete time, t is restricted to Z. Invariance of m means that m(ϕt(S)) = m(S) for measurable sets S. We perturb a dieomorphism in the 2 2 function space of m-preserving C dieomorphisms, Dim(M), and we perturb a ow ϕ by perturbing its tangent vector eldϕ ˙ in the space of C2 divergence-free 2 vector elds, m(M). For an easyX example of ergodicity, think of the torus as the square with edges identied in the usual way, and think of the ow that translates all points along lines of slope , where is an irrational number. The ow preserves area on the torus and is ergodic. There are no measurable invariant sets of intermediate area. The proof is elementary, but not immediate. Although ergodic, the ow is not stably ergodic, for the slope can be perturbed to become rational, and as in the case of the sphere, there are invariant bands that deny ergodicity. The concept of ergodicity originated in statistical mechanics. See Gallavotti’s book [Gall] for an extensive bibliography. The KAM (Kolmogorov, Arnold, and Moser) phenomenon is built on an example of an area preserving dieomorphism of the plane that has a xed point at the origin. See de la LLave’s article [de la Llave] for a thorough description of the theory. The origin is surrounded by a Cantor set of invariant closed curves on which the map is conjugate to irrational rotation. The curves bound invariant annuli, and this denies ergodicity. Enough annuli persist after perturbation in the area preserving C4 topology to show that the dieomorphism is persistently non-ergodic. It is also not partially hyperbolic, and hence does not contradict the Main Conjecture. The generalization to volume preserving dynamics in higher dimensions referred to above is due to Cheng and Sun [ChSu], Herman [Yoc], and others. Theorem E in Section 7 is the main result of our investigation of stable ergodicity that we made in collaboration with Keith Burns and Amie Wilkinson. The present paper borrows much from [BuPugShWi]. Other references appear in the relevant sections below. 2. Hyperbolicity A ow is hyperbolic if its orbits look like those in Figure 1. Orbits that are STABLE ERGODICITY 3 Figure 1. Local view of a hyperbolic orbit. The curves with arrowheads are orbits. The other lines are stable and unstable manifolds. attracted toward each other asymptotically as time increases form a stable mani- fold and those that are repelled apart form an unstable manifold. We denote the unstable and stable manifolds systematically as W u and W s. A ow is hyperbolic if all its orbits are hyperbolic. The whole phase space of the ow is lled with stable and unstable manifolds. They foliate the phase space, and as discussed in Sections 4 and 9, the properties of these invariant foliations have a lot to do with ergodicity. We denote them as u and s. Here is the formal denition. W W Denition. Aowϕ on M is hyperbolic if there is a splitting TM = Eu Eo Es such that (a) Eu, Eo, Es are continuous subbundles of TM that are invariant under Tϕ u u o o s s in the sense that Tϕt(Ep )=Eϕtp, Tϕt(Ep)=Eϕtp, and Tϕt(Ep)=Eϕtp, for all p M and all t R. (b) The orbit∈ bundle Eo is tangent∈ to the orbits, Eo = span(ϕ ˙(p)) for all p M. p ∈ (c) Tϕ exponentially expands the unstable bundle Eu in the sense that for some >1 and a constant c>0, Tϕ (v) ct v | t | | | holds for all v Eu and all t 0. (d) Tϕ exponentially∈ contracts the stable bundle Es in the sense that for some constant C, t Tϕ (v) C v | t | | | holds for all v Es and all t 0. ∈ Obviously, ϕ is hyperbolic if and only if the time reversed ow t(x)=ϕ t(x) is hyperbolic. Stability for one is unstability for the other. The leaves of uand s are the invariant manifolds, and they are everywhere tangent to Eu andW Es. W Since the bundle Eo is continuous, the orbits of ϕ are all nonsingular–ahy- perbolic ow has no xed points. The denition presupposes that TM carries a 4 C. PUGH AND MICHAEL SHUB norm since (c), (d) refer to the length of vectors in TM, but the particular choice of norm is irrelevant because we can choose the constants c, C at will. One of the basic examples of ergodicity and hyperbolicity is the geodesic ow on the unit tangent bundle of a surface of constant negative curvature. It is the mother of all examples, so we will spend some time describing it in detail in the next section.